In autonomous navigation, stable and reliable sensor data is essential for ensuring the safe and efficient operation of unmanned underwater vehicles (UUVs). This paper proposes a fault-tolerant control method for UUVs based on the Crossformer model. Crossformer is a time-series prediction model built upon the Transformer architecture, incorporating a two-stage attention mechanism to efficiently process multi-source sensor inputs over extended missions. By learning temporal and cross-dimensional dependencies, the model enables accurate reconstruction of missing or faulty sensor data. In addition, a sliding window mechanism is introduced to support real-time state prediction. The reliability of the proposed method is validated through experiments conducted on the UUV Simulator platform.